I’d imagine horse training is difficult because the horse can’t communicate how it’s feeling. Too much money invested to risk pushing the envelope or innovated training and injuring the horse. A human can communicate sore spots and adjust training and therapy to stay healthy. An injured horse and don’t really know unless it’s completely ruined.
I think you are correct that thoroughbreds as a breed are fragile compared to other horses. I don't have an opinion on whether they are more fragile now than 50 years ago. But I think we agree that thoroughbreds need not be fragile. Consider the career of Huaso:
Started as racehorse. No success by age 6; too nervous.
Next was trained for dressage. Sustained a serious injury (not apparently due to training as such) and could no longer hope for success.
Finally trained for high jump. At age 16 set the equine HJ record of 2.45 meters (1949).
Could it be that thoroughbreds started to become fragile when horses were no longer valued for military purposes?
I wouldn't say that they were fragile compared to other breeds. Most sport horse breeds are quite fragile now. Most jumping horses are contemplating retirement at age 16 and generally need a lot of veterinary intervention to keep them going. But most sport horses also have a lot of thoroughbred in them and are also very inbred.
I was talking about the fairly loose use of "too inbred." If we are looking at Inbreeding Coefficients (Wright's Coefficient) that measure relative recent inbreeding, then G1 winners (highest level) have less close inbreeding that G2 winners, G3 winners and average horses.
There are some differences when it comes to the Ancestral History Coefficient (deep inbreeding) and inbreeding to inbred ancestors.
Generally, though, you are absolutely correct - in terms of five or six generation inbreeding horses like Enable are the exception rather than the rule.
As far as the lack of diversity in the male-line, I think that has very little to do with overall levels of inbreeding as the y-chromosome carries very little genetic material.
I'm very aware of Professor Hill's work - I actually know her. With regard to the MSTN work (MSTN = myostatin which controls proportions of muscle type), there is very little conscious breeding based on actually knowledge of MSTN type (CC, CT, TT) especially in the US. Generally breeders in the US, at least, don't know the MSTN make of a given stallion (and some can be quite a surprise, running contrary to what one might have guessed from the horse's racing performance), Certainly CT horses can be found in the ranks of top class horses anywhere from 1200m to 3000m, with other factors (including biomechanical efficiency, cardio and other genes) paying a major part.
Some time on from the original "Speed Gene" work, it was also discovered that the bulky, pure sprinter type is brought about by a SINE insertion in the MSTN gene which is very commonly found in the Quarter Horse. There are CC horses without the MSTN gene that have won the Kentucky Derby.
Professor Hill has done more than one study on inbreeding. One was for the US Jockey Club, which did show that genetic inbreeding (as measured by identical runs of DNA) has increased over recent years. That led to a short-lived bid by the Jockey Club to limit the number of mares that any one stallion could cover (the Standardbred - which has far more recent inbreeding - did successfully enforce such regulation).
The other study that you mention specifically related to dna and cartilage, and that horses with a higher degree of inbreeding (close, first-time identical by descent) were more likely to have the negative variant, and less likely to run.
In terms of "modern inbreeding" that is close up paper inbreeding, Coronation V was far more inbred than Secretariat. At 8 generations her Coefficient of Inbreeding is 14.3 against 1.17 for Secretariat, so she's off the charts high, where Secretariat is low by any standard.
What is interesting is that Coronation V is inbred to Tourbillon himself an inbred horse, which would mean that she scores well on Ballou's Coefficient - the possibility that a random allele has been previously exposed to inbreeding.
In terms of Ancestral History Coefficient, Secretariat does score highly, so he is an outcross close up, but with a lot of the same strains further back.
I will note that elite horse's back from the further in time, even through to the 60s (for example) did have higher degrees of recent inbreeding, but that is because the breed was emerging from a remarkably wide background (for example Sadler's Wells, possibly the greatest European stallion of all time, is from the same mtDNA - female line - as the Shetland Pony and Icelandic horse; and there is at least one example of a cross with a Cart Horse).
Anyway, this has probably all become somewhat esoteric for most readers of this board.
The bottom line is that the thoroughbred reached what is probably the limits of performance in genetic terms about 60 years ago.
Faster times through tracks and equipment hasn't taken place (unlike humans).
Training methods haven't improved a whole lot in that time, and there is a limit to how much you could increase volume and intensity compared to humans.
The thoroughbred is more inbred that it was 60 years ago, but that isn't the cause of the stagnation.
The first reason is obviously that technology and many other aids have been introduced for human speed betterment but there is no such thing as that for horses. Also, it’s true that humans have nicer shoes and so many other things and they actually want to surpass the hurdles and improve their speed. Meanwhile, horses don’t care about this at all. It can be a reason too, right?