This all depends on what you mean by "being written". On POSIXy systems, syswrite, mtime & friends should be regarded as atomic (but probably not recovering from errors), so any OS-level write should update mtime, filesize etc - regardless of if the write is actually written to disk yet.

OS-level caching is completely transparent to these operations. Which also means that there's no such thing as "being written" - the data has either been written (given no errors occurred), or none of it has. From the POV of these operations there is no in between.

For instance:

#!/usr/local/bin/perl -w
use strict;
open F,">/tmp/test" or die $!;
for (1 .. 1000) {
  syswrite F,".";
  my $s = -s F;
  warn "$s != $_" if $s != $_;
}
[download]

The code above reports no anomalies on my system.

Confusion may set in because perl does some additional caching by itself (when using the canonical print(f) function):

#!/usr/local/bin/perl -w
use strict;
open F,">/tmp/test" or die $!;
for (1 .. 1000) {
  print F ".";
  my $s = -s F;
  warn "$s != $_" if $s != $_;
}
[download]

Gives:

0 != 1 at test.pl line 7.
0 != 2 at test.pl line 7.
0 != 3 at test.pl line 7.
[download]

etc etc.

See also syswrite.

updated to fix typo in second example

update 2: bottom line:

If your remote program sends write()s to the system often enough, and/or your peeking of mtime etc is infrequent enough, your strategy should work. But any program doing fairly heavy caching by itself (and I would guess many interesting programs do) may throw it off. So be conservative when choosing your intervals.