The current Ubuntu patch changes it from 16 to 256:

diff -uprN screen-4.0.3.orig/screen.c screen-4.0.3/screen.c
--- screen-4.0.3.orig/screen.c  2003-09-08 09:26:41.000000000 -0500
+++ screen-4.0.3/screen.c       2009-01-10 12:06:34.698560196 -0600
@@ -2026,7 +2026,7 @@ static const char months[] = "JanFebMarA
 #endif
 
 static char winmsg_buf[MAXSTR];
-#define MAX_WINMSG_REND 16     /* rendition changes */
+#define MAX_WINMSG_REND 256    /* rendition changes */
 static int winmsg_rend[MAX_WINMSG_REND];
 static int winmsg_rendpos[MAX_WINMSG_REND];
 static int winmsg_numrend;

Increasing MAX_WINMSG_REND might fix it here, but it seems to me that pushing and then popping a rendering attribute shouldn't result in any net impact on resources (other than the resulting buffer's length). I'm not familiar with the relevant code, though, so perhaps I'm missing something.

In the following:

hardstatus  alwayslastline  "%{= .r}1:Red%{= .g}2:Green%{-}3:Red%{= .g}4:Green%{-}5:Red%{= .g}6:Green%{-}7:Red%{= .g}8:Green%{-}9:Red%{= .g}10:Green%{-}11:Red%{= .g}12:Green%{-}13:Red%{= .g}14:Green%{-}15:Red%{= .g}16:Green%{-}17:Red%{= .g}18:Green%{-}19:Red%{= .g}20:Green%{-}21:Red%{= .g}22:Green%{-}23:Red%{= .g}24:Green%{-}25:Red%{= .g}26:Green%{-}27:Red%{= .g}28:Green"

'17:Red' and on will be colored green, not red as it should be.