Results

Invoking preauth filter rp_filter

No difference; all take about 2.6 ms. The same is the case for the few following stages: Short times and apparently independent of the kind of page.

Looking for node to match /acs-templating/admin/test/chain-frac-0.

Found /acs-templating/.

Performing developer support logging

Checking for changed libraries

Handling authentication

For some reason, this seems to take much longer on the Tcl-only page. Maybe because it's the first in a triple of pages that e-Tester requests? There may be a little longer time gap between chain-frac-2 and the next request of chain-frac-0

Handling authorization

An unexplained but clear distinction here: 0 is faster than 2, and 1 is slowest.

Done invoking preauth filter rp_filter (returning filter_ok)

Invoking registered procedure rp_handler

Searching for /webroot/web/brech/acs/packages/acs-templating/www/admin/test/chain-frac-0.*

Serving /webroot/web/brech/acs/packages/acs-templating/www/admin/test/chain-frac-0.tcl with rp_handle_tcl_request

Here the actual work supposedly happens. The Tcl-only page is clearly fastest. Always reparsing pages expectedly affects the templated page, and -2, which compiles two ADP pages, is affected more than -1. The benefit of -2, wrapping -1 in another include, isn't apparent; on the contrary, -1 is in all cases a bit faster than -2. The benefit of cacheing seems more than offset by the extra complexity of nesting several templates.

Invoking trace filter ad_issue_deferred_dml

For some reason, the Tcl-only page takes significantly longer.

Done invoking trace filter ad_issue_deferred_dml (returning filter_ok)

Invoking trace filter ds_trace_filter

That last phase is ended by Done invoking trace filter ds_trace_filter (returning filter_ok)

Total time

Overall, the templated pages are delivered faster. Forcing the template system to always reread all files and to recompile the ADP part slows them down, as expected, but overall they are still faster than the Tcl-only page.