The efficacy of a subchannel encoding strategy is simply a measure of how well it prevents an attacker from proving the existence of the subchannel. Of course, another desirable feature of a subchannel is the ability to embed more, rather than less, bandwidth in it. Sometimes a couple of bits of subtext are sufficient for a particular purpose; but most of the time you are more likely to need the ability to send more extensive messages. Unfortunately, the goals of bandwidth and invisibility tend to pull in opposite directions: More fiddling with bits makes detection more likely and easier.

Your first assumption in designing subchannel encoding should be that an attacker is at least as able to identify non-predictable variation as you are. Do not try to hide the message simply by assuming an attacker will not know where to look. The key in maintaining the invisibility of a subchannel is to make sure that the distribution and pattern of subchannel bits closely match those in a typical file of the same format.

In many cases, the expected distribution of pre-encoded subchannels will be uniform and stochastic -- but not always. You have to look at whether there is a bias toward 1 (or 0) in the pre-encoded subchannel slot (the bits or variations you have identified as encoding sites); but you also have to look at whether there is a frequency shift between the start and end of a file and/or whether cyclicalities exist in bit frequencies of pre-encoded subchannels. A good first step is to extract a large number of pre-encoded subchannels, and see if this data is compressible (if so, it is not purely stochastic and uniform, and you need to look more closely at the patterns).