Headthump: I shouldn't even dignify a Half-Life mapper's claims with a response, but if you insist.
His argument is based on tests that don't have anything to do with real life mapping. In his first, he shows a light inserted within a 'pole', and somehow claims this proves that the bsp splitter is god's gift to render speeds. Without showing the tris, I see little reason to believe his argument. Here's the same piece, done in q3, on the bottom showing an unmitred peice, the top mitred.
http://scampie.spawnpoint.org/d3/q3_mitre_compare_2.jpg
If he's in fact not lying about there being no cut there, then there's more to the story than we're seeing. Possibly a decal? I know they exist in HL, but don't know how common they are in level design.
His second piece of 'evidence' just shows an example of the splitter doing it's job correctly and how we'd all expect it to work. A real world example, like in my first Q3 mitre comparision (
http://scampie.spawnpoint.org/d3/q3_mitre_compare.jpg ), shows the compiler just doesn't how how to effeciently deal with all the vertices of the various faces coming together and has to split things up. By mitering, and making more vertices meet at common places, less splits occur and thus less polygons are created.
Mitred brushes making the map cleaner is a personal call, I find in my mapping where I'm very concerned with overall design, it's much easier to see things. As for his final point, plane counts are a Half-Life compiler specific concern, and may well be true for all I know.
That's all I have for now.