The important part is that the anonymity set - the amount of traffic among which those with a need for anonymity can hide - becomes larger no matter what the people without a need for anonymity do.
Consider the extreme case where Tor is only used by dissidents in a single country X. As soon as country X's secret police observes your home internet connection connect to Tor relays, they know that you are a dissident.
Now consider the case where only 1% of the traffic on Tor is by dissidents in country X. When your home internet connection is observed to carry Tor traffic, it is impossible to tell whether you're among the 1% of dissidents or the 99% of non-dissidents. So they have some reason to suspect you, but it's already a clear win, because rounding up and terrorizing 100x as many people takes more effort and is more likely to result in pushback.
The only place in the Tor network where the "trying to remain anonymous" makes a difference is when the secret police collects exit relay traffic. However, if all they see is that 1% of exit traffic is TLS sessions to dissidents.xx and the other 99% is unencrypted sessions to facebook.com (hypothetically), that still doesn't help them figure out which Tor clients are sending those 1% of traffic that they want to chase down.
Of course, all of the above is subject to the inherent limitations of Tor (e.g., somebody who is able to observe all relays can do statistical timing-based attacks to correlate relay input-streams with relay output-streams; they can then trace back the Tor circuits and figure out which user is responsible for which exit traffic; somebody who is able to observe a fraction of relays will be able to do such correlation attacks with a certain probability of success; the secret police might observe dissidents.xx as well as the home connections of everybody using Tor, and might be able to sieve out the 1% of dissidents using timing correlation, etc. [0]). The point is that the nature of the 99% of non-targeted traffic doesn't matter; the important thing is that it's there, and the more, the better.
[0] This seems to suggest that if you want to hide some of your traffic via Tor, it actually makes sense to tunnel everything via Tor. However, this also has problems: for example, if you use the same browser to access both facebook.com and dissidents.xx, browser fingerprinting might kill you. I don't actually know what the best practices recommendation is. Given an adversary with sufficiently tight control over the communications infrastructure, you're basically screwed.
Consider the extreme case where Tor is only used by dissidents in a single country X. As soon as country X's secret police observes your home internet connection connect to Tor relays, they know that you are a dissident.
Now consider the case where only 1% of the traffic on Tor is by dissidents in country X. When your home internet connection is observed to carry Tor traffic, it is impossible to tell whether you're among the 1% of dissidents or the 99% of non-dissidents. So they have some reason to suspect you, but it's already a clear win, because rounding up and terrorizing 100x as many people takes more effort and is more likely to result in pushback.
The only place in the Tor network where the "trying to remain anonymous" makes a difference is when the secret police collects exit relay traffic. However, if all they see is that 1% of exit traffic is TLS sessions to dissidents.xx and the other 99% is unencrypted sessions to facebook.com (hypothetically), that still doesn't help them figure out which Tor clients are sending those 1% of traffic that they want to chase down.
Of course, all of the above is subject to the inherent limitations of Tor (e.g., somebody who is able to observe all relays can do statistical timing-based attacks to correlate relay input-streams with relay output-streams; they can then trace back the Tor circuits and figure out which user is responsible for which exit traffic; somebody who is able to observe a fraction of relays will be able to do such correlation attacks with a certain probability of success; the secret police might observe dissidents.xx as well as the home connections of everybody using Tor, and might be able to sieve out the 1% of dissidents using timing correlation, etc. [0]). The point is that the nature of the 99% of non-targeted traffic doesn't matter; the important thing is that it's there, and the more, the better.
[0] This seems to suggest that if you want to hide some of your traffic via Tor, it actually makes sense to tunnel everything via Tor. However, this also has problems: for example, if you use the same browser to access both facebook.com and dissidents.xx, browser fingerprinting might kill you. I don't actually know what the best practices recommendation is. Given an adversary with sufficiently tight control over the communications infrastructure, you're basically screwed.