I have had this problem as well.  My box is a Thinkpad 600 (233 MHz)
with
  com0 at isa0 port 0x3f8-0x3ff irq 4: ns16550a, working fifo

It runs 1.6.1 and pppd with a 28.8k modem at 57600 on the serial port
and a USB ethernet.  Most of the time I get very few errors, but when
I left myself logged in to the console (X11) I got a lot.  Typical
errors are:

com0: 6 silo overflows, 0 ibuf floods
com0: 4 silo overflows, 0 ibuf floods
com0: 4 silo overflows, 0 ibuf floods
com0: 11 silo overflows, 0 ibuf floods
com0: 16 silo overflows, 0 ibuf floods
com0: 1 silo overflow, 0 ibuf floods

which cause PPP FCS/framing errors. 

I am fairly sure there is no MP code in 1.6.1, but that doesn't mean
no changes were made for finer-grained locking.

Are you running X?

It would be cool to have an interrupt priority level trace facility
that logged levels and cycle counter times.  This would pretty clearly
have to be a special compile of the kernel to replace/adjust the spl()
calls and catch interrupt-routine entry/exit.

        Greg Troxel <gdt@ir.bbn.com>