To analyze this we need to think carefully about various scenarios that stipulate what knowledge the attacker has, and then reason about what other information they can deduce.
Mark Burnett's answer already provides us with one such scenario: if the attacker acquires your password from one of the sites, then they know all but two characters of your password in the others (taking your suggestion of using "a couple" salt characters literally), and thus can easily guess the others. Since the point of using different passwords in different sites is to protect you against an attacker who learns your password in one site, this pretty much shoots your proposal down already, I'd say.
Here's a second scenario: the attacker doesn't know any of your passwords, but they know that you might construct them according to the rule you've described, which I'll spell out like this:
- Construct a shared, master password according to some general rule;
- Construct a variant sub-password for each site by appending a two-character salt.
If the attacker knows that you create your passwords like this, then strategies for guessing #1 adapt very easily for guessing #2, because the two-character salt barely adds any entropy to the master password.
Third scenario: an attacker who doesn't know any of your passwords, nor know or guess that you construct them according to this procedure. Then I'd say that whether it's effective depends on what strategy they use to guess passwords. You'd have to know or guess their password guessing strategy, which you don't. But then have no reason to believe your password scheme will be any more or less secure than an alternative.
The only virtue I can see to this, then, is that it makes your passwords easier to remember. But I think the first scenario shoots the whole approach down, period. And the sad fact is that things that make passwords easier to remember generally make them easier to guess as well.