"It's faster to just use one mysql table with 300 columns then linking the data in 10 tables. 10 tables mean 10x the load time!"...

- External consultant hired for optomizing database. (He didn't make it through the day and was sent home...)

technically true (for reading), it is however a hell of a lot slower to update if some of the data is shared between rows and needs to be kept in sync (since you might have to write thousands of rows to change a single value then).

If your normalized data is too complex to read quickly you should probably create separate denormalized models for reading. (doesn't even need to use the same database)