Extreme Modification index

#1
I am working on Confirmatory Factor Analysis in LISREL. Now, my first model reports an extremely high Modification Index (MI=2498,71) for one specific Lambda-X parameter.

I decided NOT to free this parameter, but to free a parameter for associated measurement error between two items that were involved in the factor associated with the extremely high MI. This would be in correspondence with earlier results from another sample - were those two items had been a problem as well. This resulted in a model with significant better fit, and the MI for Lambda-X of concern in the second model was reduced from 2498.71 to 44.43. So, this seems to be a reasonable choice, resulting in better fit, with much smaller strain in the model.

Now, my questions is, should i be concerned with this initial extremely high MI? Somehow i remember reading somewhere one should take this as a serious indicator something else is wrong, or it is a very instable solution. But i cannot find it in any book anymore. Is it an indication of spurious results or other problems?