As an extension to the question posed recently by Bulatov, I wonder what are the maximal sub-classes of perfect graphs for which we know of combinatorial algorithms to compute a maximum independent set.
One springs to mind and is listed as a maximal subclass in ISGCI, which surprised me: perfect claw-free graphs (a.k.a. perfect quasi-line graphs). This was done by Minty for all claw-free graphs around 1980. But a couple of other algorithms for claw-free graphs, one recently in SODA 2011 that is $O(n^3)$ by Faenza, Oriolo, and Stauffer, use the Chudnovsky-Seymour structural characterization of these graphs to reduce the problem to line graphs (and therefore maximum matching) in a fairly straightforward way. If you're only looking at perfect claw-free graphs, then the earlier characterization by Maffray and Reed is sufficient (and the reduction to line graphs is more obvious).