I just opened a PR on this.  

I *suspect* that the workaround for the "Web load causes reboot" is to
turn off pseudo-device rnd.

So, what's going on here is that rnd.c assumes that splsoftclock()  is
like other splfoo() calls in that it masks off interrupts.

This is not, in fact, the case.  It *enables* interrupts, and thus
*un-protects* a chunk of code from interrupts.. in this case, against
a recursive softnet() interrupt.

					- Bill