weird serial console problem

[Valery Ushakov <uwe%stderr.spb.ru@localhost>]

I have encountered a weird problem with the serial console on my
landisk.  Current current, shell, no line editing:

  $ cat runme
  #!/bin/sh
  d='.123456789'
  /bin/echo $d$d$d$d$d$d.1234     # plus \r\n
  /bin/echo -n $d$d.12345678
  sleep 10
  $ ./runme
  <... no output ...>
  .123456789.123456789.123456789.123456789.123456789.123456789.1234
  .123456789.123456789.12345678$ 

when the script is run, nothing is displayed unless some additional
output kicks the console back into life (the shell prompt after the
sleep, typing a key, writing to constty from another session).  The
echo has actually completed successfully by that time, but its output
is not printed.

This is actually a reduction of a bug that I encountered in the wild

  <uwe@felix:~> (1135) id
  <... stuck until more output ...>
  uid=1000(uwe) gid=1000(uwe) groups=1000(uwe),0(wheel),5(operator)
  <uwe@felix:~> (1135) 

that emulates the output of the program and the bash prompt (including
invisible bracketed paste escapes that readline now uses by default).

The "runme" reproducer actually succeeds some times, but it's probably
b/c there's an extra byte somewhere in the output path that breaks the
numerology, or something.  The id(1) command gets stuck 100% reliably,
as well (as luck would have it) as "ls -aCF" in ~ :) (before I added
"runme" and broke the numerology, the -F that adds some slashes is
important to get the numbers right too).

I didn't study the numerology in more details.

I can't easily test a different USB/serial adapter, as the DB9 serial
cable a friend soldered years ago has failed recently and he replaced
it with a uchcom dongle.

  uchcom0: QinHeng Electronics (0x1a86) USB Serial (0x7523), rev 1.10/2.64, addr 4


Does this ring a bell?  Since I've never seen anything like this
before I don't think a lurking bug in scif(4) is likely, and given
that the serial console change was very recent it makes me suspicious.
But is this kind of failure mode even plausible for a usb-serial
gizmo?  I'm too hw-illiterate to have any clue...

-uwe