What have you used to do the PDF conversion? I wonder if Adobe would work? They also make several products to convert a print spool to a PDF such as CutePDF. I am not sure if those create an image PDF or a text PDF, but if it is a huge word document that you are not able to work with any other way, it may be worth a shot.
PDF writers can be notoriously variable in the way they construct things internally.
These days the Adobe products appear to be quite consistent and the latest versions (say in the last year or so) of some of the third party products seem to produce more good and consistent files for re-interpretation than they used to.
At the same time Monarch is being tuned over time to address the variability of the existing writers as they are identified and analysed.
I assume you have experimented with the settings that can be adjusted to improve the consistency of the Monarch interpreted output. From time to time I have found some reports (or variants) which are not themselves consistent from run to run, or even page to page! My approach then has been to find a way of modelling to grab all the required data in whatever way I can and then deal with the remaining issues using calculated fields.
If the original report is not too confidential to share with others you might consider asking others to have a look at it to see what results they can obtain - it is a sort of 'black art'.