I had a short look at that.
26 seconds is the boundary where trouble starts (updates more
than once a second) Does 256 ring a bell here? The function
used is "timeout(millisecs)" from curses. The manual mentions
no restriction on the number of millisecs supported. The limit
is likely to be around 25600 millisecs as 26000 seems to fail.
A short look into  lib/libcurses/timeout.c  shows that it  is the
VTIME parameter(char!) of the termios interface - thus the systat
implementation will currently be able to sensibly sleep 1-25 secs.
For the curious: Above that the sleep time wraps to <secs-26>.4 s
for 26-51 secs then +0.8, +0.2, +0.6, +0.0. But that is not useful
unless you want the fast updates :-)

Looks either like a doc bug or some more SMOP to get it o match 
users expectations.