There is no real consensus on gap penalty, because the 'best' gap penalty is dependent on which sequences you want to align. If you want to align proteins of two very distant organisms you want to set the gap penalties lower than if you want to align two proteins of closely related organisms.

As for the default value of ClustalX, according to this article on Effects of Gap Open and Gap Extension Penalties ClustalW uses GOP 15.0, GEP 6.66 by default. GOP is the gap opening penalty and GEP is the gep extension penalty. I think ClustalX uses the same default as ClustalW.

I can't find it on their website right now, but are you sure that ClustalX uses PAM250? It's an outdated matrix and I would imagine ClustalX to use BLOSUM.

[Clarification]

The article presents a demo alignment of 3 sequences with length 4, 3 and 3 acids as follows:

ACGH
CFG
EAC,

which aligns both in the article and Clustal as:

-AC-GH
--CFG-
EAC---

But if I use the implied gap penalty value of 8, I get:

ACGH
-CFG
-EAC

It is possible in my implementation to get to the "article demo"/Clustal alignment, but for that to happen I have to change the gap penalty from 8 to 5. So my question refines to "No matter how outdated the PAM250 is, what is the bioinformaticians' consensus on the gap penalty value when dealing with the aforementioned substitution matrix?"