Edward Tufte, in The Visual Display of Quantitative Information (1998, 2nd ed. 2001) - from the aricle you linked to - is completely opposed to perceptual scaling, and I am not worthy to argue with him. However, it does depend on what you are trying to do.
If you are trying to clearly represent data accurately to a user, then you should use real sizes, because otherwise you are misrepresenting the data. You may have to provide more information about the data, and ways of interpreting it to help people clearly visualise it.
If, however, you are not concered about the data accuracy, but wish to give an impression of comparative values, then it might be appropriate. But you should also then be clear about what you are doing.
Tuftes problem is that if you do not represent the data accurately, you can easily overemphasise things that you wish to - in this case, differences. If A is twice the size of B, and this relationship is critical, the A should be displayed as twice the size of B. If it is more critical that users see that A is verys significantly larger than B, it might be appropriate to emphasise this. But when people accuse you are misrepresenting the data, have an answer ready.