I think it's a case of a paper's authors being pretentious. I went digging for some examples, the best I found is this one: http://books.google.com/books?id=X88_R8gH4hsC&lpg=PA54&ots=-FLjG-dNZg&dq=%22fractional%20algorithm%22&pg=PA54#v=onepage&q=%22fractional%20algorithm%22&f=false
The paper writes:
...we show a fractional algorithm for the switch throughput problem, i.e. one that can insert fractions of packets* [...] Then we transform our fractional algorithm into a discrete algorithm, i.e. one that can insert and transit integral packets.
My understanding suggests that a "fractional algorithm" is one that can process sub-integral, but not necessarily continuous (i.e. "a stream") units of data. Obviously this only applies to certain classes of algorithms, but an example could be an image-processing algorithm: a fractional approach might be able to work on an arbitrarily sub-pixel basis rather than per-pixel (i.e. discrete units), but it couldn't necessarily process a stream of color data (e.g. an analog TV scanline).