Aha! I found it!
Turns out it is using a vector magnitude to get the max speed, so the square root of (a^2 + b^2 + c^2). That's why you had to keep the slide speeds so low. Well, this is sort of not really a bug but an unintented consequence -- I may have to ask the team about this... I'm sure most of the time max_speed is supposed to only involve Z (forward/back), but are there situations where we'd want it to involve sliding?
In the meantime, until this 'bug' is fixed (if it will be), you can do this: increase the slide speeds and lower the forward speed, since technically the forward speed is ignored in favor of the magnitude of the whole. If you want the slide speeds to be half the official forward speed, if I'm doing the math right: N = sqrt((Z^2)/2), where Z is the 'official' forward speed, and N gives you the new speed to put in the table.