Make the ad*.mp3 files much smaller.
    In particular, Steve Snyder remarks that squid's default
	maximum_object_size_in_memory 8 KB
    may preclude this Adzapper file from ever being accessed as TCP_MEM_HIT.
    And anyway there's no need for it to be 1 second long. That was just
    the smallest ready made silent MP3 I found on the web.

Adzapper performance tuning.
    How many zappers should one run in parallel?
	On low RAM machines many zappers can be a real lose, and even
	when RAM isn't an issue, since every URL must pass through the
	zapper before Squid dispatches a real fetch the parallelism may
	be a net lose anyway.
	My current rule of thumb is "as few as possible before the squid
	complains about ``all redirectors being busy''". Web traffic is
	bursty, and generally fractally noisy (i.e. the spikiness is
	pretty much the same at all granularities) unless the upstream link
	is clogged. So more zappers may not help much.
	The only real likelihood of gain is if squid has a serial
		get request, redirect, dispatch it async
	main loop. In which can busy redirectors really would be bad and
	many redirectors is indeed a win. But I haven't read the squid code
	in this regard.
    Does coalescing the adjacent patterns of the same type help or hinder?
	I've seen anecdotal remarks that this can be a lose.

Join some of the ad pattern mailing lists that are assoicated with some
    of the other zapping programs and their maintainers.

Paul Weber's "check my zapper is up to date" idea.
    Magic URL/CGI to check zapper version against current one.