"I think I read an article that compared a few different password hashing strategies, I'm interested in your thoughts..." (and then follow up with a good comparison that makes the point)
"Have you compared X, Y and Z?" Tone is important with this one, but it can be the most straightforward.
"At the last place I worked, the paranoid guy insisted that we do X because he thought it was better -- it was just as easy to implement, is that something you'd be open to?"
When all else fails, you can follow up in private!
I think the key in any of these exchanges is to offer new information in a way that doesn't question the judgement of the other individual. Basically, assume that everyone is rational and differences in judgement are due to differences in information/experience. Then, when someone makes a judgement different from yours, you are merely giving them the information that lead to your decision -- but leaving them to make up their own mind. Most people are more ego-identified with their judgement than with the information they have.
Welcome to my quotes file.
Bad: This variable name goes against our naming conventions. Please change it.
Good: Why did you decide to go against our naming conventions here? Did they hamper readability?
Not only does it keep the other person from becoming defensive, you might actually discover that they have valuable insight.
I find questions most useful when the issue really is a matter of opinion, and they're just as likely right as you. For example, "why did you decide to key this webpage's data by content hash instead of by URL?" This is a sincere question - there may be arguments for both approaches, and the point in asking is to tease out those arguments and figure out what actually is the truth. It's an acknowledgement that you might be wrong, and not an attempt to paper over the certainty that you're right with diplomatic phrasing.
One trick I've used many times to correct someone without embarrassing them is to couch my own feedback in conditionals, something like "You've been studying this a lot more than I and I'm sure you're probably right but I seem to recall reading a bit ago that an md5 sum isn't a sufficient way to hash a password, you need to do more processing."
That then gives them an out "yeah I looked into that and I'm pretty sure this is best practice but I'll check again" or somesuch. Probably 80 percent of the time, when I frame it like that, I'll see a followup saying that they've gone back and discovered the right way and will be doing it that way.
The really nice thing is that people are appreciative when you treat them that way, and far from creating a new enemy you'll have earned respect and trust from that person.