Once a team starts to work overtime and weekends to finish stories (on its own accord or by instruction) then the value of your velocity measurement is shot to pieces. Predicting and committing to future sprints is then a complete guesswork (or rather even more of a guesswork..) when bad committal is masqueraded with pulling a rabbit out of hat.
It also masks the problem in the retrospective - in my view the most important part of scrum. Developers put down positives for going above and beyond rather than being forced to admit, and therefore analyse, that the team failed.
I've found, though perhaps it is a British cultural thing, that there are plenty of things people sit on and don't mention out of politeness - they only come out after a particularly bad failure.
