No one ever said of a defeated commander "At least he still had some troops in reserve".
The only really rational way to look at it is to try to really honestly evaluate whether there was a chance of throwing in the last reserves turning the trick - whether they got thrown in or not - and whether what they could have been doing instead would be preferable. Still having unblooded reserves can often let you mitigate the consequences of a lost battle, for example.
With hindsight, we can argue that Napoleon could perhaps have won the overall Waterloo
campaign if he'd pulled back on the 18th with his Guards to cover the movement, replanned, and perhaps marched by his left (to pull the Allies further from Prussian territory, perhaps). Or we can argue that, even though throwing in the Guard failed, it was a choice that had a reasonable chance of working - on the grounds that if the British army had been broken (and certainly Wellington thought it was a near run thing) he could have salvaged the campaign.
One has to factor in the random element, too, of course. 80% chances fail one time in five, but the ones that happen to fail are still
better choices than the 20% chances that happen to succeed...
In this light I think Lee sending in Pickett's Charge (for example) can be evaluated in terms of risk-reward and what else he was going to do with the troops. It came off badly, but if it had been successful it would not perhaps have looked surprising on the
tactical scale (attacking a previously bombarded area of the enemy line with significant strength) while having a significant impact on the campaign; there were no serious consequences beyond the casualties (which were painful); Meade did not make a counterstroke.
You can argue either way.
For Hooker in the late part of Chancellorsville, there is perhaps the opportunity to do some damage, but it would be surprising on the tactical scale (the attack has to go through bad terrain) and there would be serious consequences for failure besides the casualties (as we know Lee was already willing to launch a counterstroke).
Perhaps it might have worked? But the probability-adjusted risk is too high for the reward; going back and trying again another day is a better option.