em driver - issue #2

[EM1897 at aol.com]

Some more info on this:

netstat -m, after the "lockup" of the em driver shows:

533/5533/80000 mbufs in use
532/1532/20000 mbuf clusters in use

So there really shouldn't have been any failure to get a replacement 
buffer in the first place. The max clusters is interestingly exactly
1000 clusters over the idle amount, and mbufs is exactly 5000
more, which can't be a coincidence. Could there be some sort of 
threshold in play here?