Assuming the regen is at least linear with speed, that is there's not an increase in generator efficiency at high speed, then I'd think that the the more regen and hence the slower you went the better.
There's a fixed amount of potential energy available from the loss of altitude, Mass x Height x Gravity's Acceleration, so the question is just where does the energy go. If you don't use the brakes, the energy loss is mostly air resistance, rolling resistance, and regen. Rolling resistance doesn't vary much with speed, but air resistance goes up with the square of the speed, so my guess is that the best strategy from the standpoint of maximizing charge is use as much regen as possible and go as slow as you can stand.
That would mean pretty slow speeds though, so to be practical, I'd suggest something like stay below 50 or so using just regen to control speed. Below 50, the air resistance is still reasonable, but because of the square law, the loss to air resistance starts getting pretty bad as you go much faster. At 70 almost twice as much energy will be lost to air resistance as at 50 (490 vs 250). Because of regen losses though, if you want to be going faster at the bottom, as abasile said, speeding up by backing off regen seems like a better strategy than using the motor.
Good stuff, RDoc. I've never yet found a highway grade steep enough that full regen was needed to hold a constant speed on the downhills; at most, I've seen something like 30kW of regen (and usually less) with the cruise control set at 65 mph, while coming down the big mountain passes here in CO. I've read elsewhere on TMC that the standard regen algorithm is tuned to yield a deceleration of 0.1g, with a cap of 60kW, though my empirical observation is that it's more complex than that. If we assume no friction and no aero resistance, a gravitational acceleration of 0.1g is achieved on a slope of about 5.73 degrees, also known as a 10% grade. But aero resistance at highway speeds is quite a large factor, so a highway grade would have to be a fair bit steeper than 10% before full regen is required to hold a constant speed. There's a tipping point, though: at higher speeds and depending on vehicle gross weight, 60kW of regen is no longer sufficient to give 0.1g of deceleration.
Where am I going with this? Beats the hell out of me...but I do agree that for longer, steeper slopes, using regen to hold a reasonably low constant speed ends up giving you more range than using regen to hold a higher constant speed, and even more so compared to freewheeling down the mountain at terminal velocity.