The MFI method has underperformed in the recent decade according to Mungofitch I think...

Poor performance after a model is build is pretty normal for any quant method...backtests are never like reality.
The hope is that a method that backtests at market+10% might get you market+3% in reality.

The problem with the "magic formula" is a bit more interesting, and a bit bigger.
The method he proposed simply does not perform as he says it does, in the exact historical period that he originally tested.
As in: not within 10%/year of what he claimed, which (IIRC) was in the vicinity of 28%/year.
This was tested by quite a number of observers, with some testing databases of outstanding quality.

There are various possible explanations for this gap.

* His testing database or software was not of sufficiently high quality.
* He used some selection criteria that were not disclosed in the book.
* He made some errors in his testing methodology, including the risk of substantially overtuning to his test data set.
* He fibbed.
* Something else I haven't thought of?

He's a pretty respected person, so I personally do not suspect #4.
More likely some mix of #2 and #3.

Bottom line: the issue is NOT that MFI has not done well since it was published.
The issue is that it didn't do well (as described) before it was published.

Jim