lets do the math with your numbers.
hundreds of boxes: at least 200
a crash every other day: say 200 crashes a year.
if 200 servers have a total of 200 crashes collectively
that would be one crash per year per box. hardly every few days.
when you take an average of one year uptime per box and refer to it as crashing every few days.
sounds like crap and stupid to me.
Let's put it another way. The Windows shop has to deal with a critical failure almost every day. I'm talking about a failure which impacts company productivity.
The same thing happens in the Unix shop about once per year.
Both shops being otherwise equal... same number of servers, admins, etc, which would you rather support?
You can play with number all you want, but you can't bury the fact that Unix is far more reliable than Windows.