by 'low latency' they mean two things :
1) 'efficient' use of data transmission capacity, i.e. whether chaff is sent(expensive) or not.
Chaff might be really only "expensive" if 1) Monetary, user chose to pay for it under metered plan, or refuses to buildout free p2p, guerilla, mesh networks. 2) Bandwidth, if chaff does not get out of the way upon demand by wheat for carriage. 3) Footshoot, user attempted to feed a higher committed rate to their CPU or internet than either can handle.
2) actual low latency. In order to prevent timing attacks, packets need to be reclocked, which means adding delay, which results in higher 'latency'.
While reclocking is likely necessary part of TA defence, many CPUs and NICs can handle line rate processing, and the committed rate to an overlay network can be set below both the physical link speed, and the rate bought from the ISP within which the overlay rides. User or overlay can create its own processing headroom by choosing a lesser rate. Also, depending on nature of input, reclocking may not necessarily imply additional average delay, as packets and gaps between them might be simply normalized. randomized and or distributed within the same overall sum. Sure maybe due to variety of hardware making up an actual overlay network and users paths in it, and other processing demands, latency may be higher than non chaff network, but maybe not by enough to actually preclude use of low latency apps such as voice chat, IRC, etc.
any low-latency web onion router - could not defeat The Man
This seems yet to be lacking proof and perhaps cannot actually be said without it.