SUP will always have more overhead for a single feed, because it requires more than one (conditional) request: you need one for the SUP feed each time it is pol

SUP will always have more overhead for a single feed, because it requires more than one (conditional) request: you need one for the SUP feed each time it is polled, and one for the content feed whenever it is changed. Even with 10 feeds, SUP will probably not break even because processing 1 conditional request for the SUP feed is not really less work than processing 10 conditional requests. Each way will have approximately the same network latency (you can process 10 GET requests in one round-trip with pipelining) and the same bandwidth usage (you can stuff a few GET requests and a few 304 Not modified responses in a single packet with pipelining). The same techniques used to efficiently update a SUP feed can be used to keep individual cache entries for each individual feed up to date. Even without pipelining, the benefits are of SUP will be negligible for a site with just a handful of feeds. The fixed cost of implementing SUP and the variable cost of maintaining it needs to be included in the cost analysis, not just the bandwidth. SUP and Sixapart's stream will be useful for big sites. But, they are unnecessary for small ones, at least as far as they've been documented so far.