> Could it be the usual 'worse in the evenings' scenario?
Rather the reverse in fact or at least "no pattern". If one forces a resync by s/w-rebooting the DSL router, and this could be in the late evening, that's when you could get a 544k US sync for example. The other night, s/w-rebooted the unit (which I'm assuming does an orderly protocol shutdown as from experience this seems to be the way to keep BTW happy) and got 544 US which then lasted until about 10:30 next day.
[Still haven't managed to capture a "before" =high US sync bit loading data set, but I will get there.]