I've run the same brushless motor and controller on both 36V and 48V on the same bike. What I found was that the low end torque seemed to stay pretty much the same for either voltage as far as I could tell, top motor-assist speed improved as expected and mid to top range torque also increased slightly (there was a noticeable improvement in throttle response when accelerating in the mid-range). Cruise speed current was less on 48V than on 36V, but there was a temptation to just cruise faster and so drain the battery more quickly, as Nick has mentioned!
The usual limit on controller upper voltage, at least for controllers in the 24V to 48V range, is the voltage rating of the main capacitor(s) across the supply. The other restriction is that some controllers have fixed low voltage cut off point for their set voltage, which means that some 48V controllers won't work at 36V, because the controller interprets 36V as a flat battery. This is the reason I have to run my Tongxin at 36V when 24V would be fine for my purposes, I think. The Tongxin controller cuts off at 29V and can't easily be adjusted as it's potted in resin.
If going over 48V, then controllers often need output stage FETs with a higher voltage rating and also occasionally need the internal voltage regulator modifying to accept the higher input voltage, in addition to changing the main capacitors for higher voltage ones. Most non-resin potted Chinese controllers are pretty easy to tweak for use on higher voltages and the current limit can usually be fairly easily adjusted as well, although the output FETs and size of the main capacitor(s) may need to be changed to cope with the greater current (the capacitors work hard when absorbing the turn off spike current and supplying the peak current each time a motor phase is commutated).
There are a few excellent threads on modifying and repairing various types of controller over on the ES Forum if you would like more information.
Jeremy