Note the OP is asking a "theoretical question" -- and I said "*estimated* marathon pace". Of course, with actual times we could be more precise, but this is a theoretical exercise without really much background information to go on, nor well described constraints. Without a recent marathon time, you need to start with a reasonable "estimate" based on the recent time given for 10K.
For a 36:50 10K, McMillan gives a range of "stamina paces" as fast as 5:54 for "Tempo Intervals" (or even a faster 5:48 for Daniels-style "Cruise Intervals") to as slow as 6:31 for longer "Steady-State" runs. 6:30 pace is just at the slowest end of this "stamina" pace range, and estimated MacMillan marathon pace is a slightly slower 6:36. Daniel's estimates a slightly faster marathon pace pretty much exactly at 6:30.
There are many ways to estimate paces, but they won't change that much between the most common methods. I only use "marathon pace" to give perspective of how slow this explicitly selected "tempo" pace is. If the theoretical runner actually improves his stamina with this single training at 6:30 pace, not faster or slower, then this training pace will become proportionally slower, like his estimated 50k pace or 80k pace.
I agree with "malmo" that "tempo" isn't a single pace to hit, but can be as slow as marathon race pace, or as fast as 10K race pace, depending on which stamina workout you are doing, and a few other factors. I think of "tempo" as a synonym for "rhythm" rather than linking it to any pace, and prefer to speak about training in result oriented terms of improving target variables like stamina and speed and endurance.