Great update

End stops

A servo installation should not be hitting mechanical end stops.  I'm surprised the heatsink gets hot instead of the servo burning out.  Did the servo buzz while in this condition?

The servo arm position and linkage should be adjusted in conjunction with endpoint values set in software such that the servo motion is not stopped by a mechanical limit.

The whole point of using a servo is to set the position electronically and hold that position.

When you say you "remove the drive", is that stopping the stream of pulses or removing the power to the servo?  In both cases you are relying on the friction in the servo gear train to hold the position.

If the repeated update pulses are removed from the servo the electronics in the servo are sitting there waiting for a pulse which seems to be an invitation to noise glitches.  Also if some force moves the servo arm away from the set position there are no pulse updates to drive it back.  If this method must used the output to the servo should not be left in the high impedance state but driven to logic low to reduce noise susceptibility.

If the solution was to remove power from the servo the noise susceptibility is removed but the ability to correct the position is also removed.

Not so much a stall as a pause

Thank you, Ted , for the comments.

What was actually happening (and I didn't go into details since I was trying to keep the post brief) is that, during adjustment by the user of the servo movement end limits, some users were pausing for an appreciable time when the servo had driven the point rail to meet the stock rail - probably to read the setup instructions about the next step - or were even continuing to try to move the servo further towards (beyond) the end limit.

Since I had not anticipated that this would happen, I did not 'detach' the servo (ie. stop the stream of drive pulses) between each manual adjustment of servo position. The servo, therefore, continued to draw an appreciable current through the QSDD regulator which would then be trying to dissipate several watts of power via its small heatsink - and got too hot.

The change in the software was simply to 'detach' the servo between each manual adjustment, stopping it taking any current, and therefore keeping the regulator cool regardless of how long the user took to complete the adjustments.

In normal operation, the QSDD 'detaches' all servos after the required movement is completed and, as you say, you may then rely on the friction in the gear-train to hold the servo in position. This has not been reported as a problem by anyone (so far) although I should say that, in my own case, I use Peco turnouts which have an over-centre spring to hold the points positively in postion and do not rely on the gear-train friction.