It's a non-standard (NoSQL) database. Even SQL databases, despite years of standardization and efforts at poaching each other's customers, still have rough edges that make moving between different SQL database products non-trivial. Just because you can get your data out doesn't mean you're not locked in to all of SimpleDB/DynamoDB's quirks, of which there will be a lot more because it's not following a standard approach. Your code will have to go through contortions to work around / adapt to the SimpleDB limitations; effort which might well be wasted or counter-productive on a different system. That's the lock-in with AWS in general: API lock-in, not data format lock-in.

DynamoDB and SimpleDB are also a lot simpler than SQL databases. As far as I can tell, DynamoDB is a key/value DB with support for ranges and MapReduce, and not dissimilar to other NoSQL databases like Riak.

There may be instances where large datasets are hard to migrate from DynamoDB, but overall it doesn't look to me like lock-in would be that much of a problem, assuming you have a decent abstraction layer.