Hello Björn,
apologies for the late reply. I've been thinking about this and I can suggest one thing that might help you out. In the Silverlight version there is an "ad-hoc" facility which produces a coding report in JSON format. This is specifically designed to do computer-driven analysis (currently used by two separate side-projects of ours) and, if you're happy to use the Zotero API, it's possible that you'll be able to "digest" the JSON reports and feed the result to the API itself.
Here is how to obtain a JSON report:
First, you'll want to obtain the list of references you're interested in, I'll assume you have your ways to get hold of this list. From the main "Documents" page, you would then select all documents in your list and click on the "Get coding report(s) for selected item(s) that have been completed". It's in the main toolbar two buttons on the right of the "printer" (Export bibliography) button.
You would then tick one or more coding tools (codesets) to indicate what coding data you're interested in, tick the "Export results as JSON text file" option and click "Get report(s)".
The resulting file starts with a "CodeSets" which represents the structure of the codesets in question, including name, description, IDs and their place within the coding tool tree. The file then includes a "References", which lists all the references that were selected, in the "Codes" field, you'll find a "flat" list of codes (from the coding tools that were selected) that have been applied to the reference in question. The Codes list includes an AttributeId which is used to "find" the code (and it's place) in the "CodeSets" section, as well as all data associated with it (including the text associated to the coding, from the full text-document, if any; field is "ItemAttributeFullTextDetails").
One word of warning: should you decide to use this facility, I must stress that it's quite possible that the JSON format might change (not dramatically) without warning. Likely changes are the addition of more fields, it's unlikely that we'll change the name or position of existing fields. This means that you should be careful when writing your own parsing solution: it should be designed to cope with the appearance of additional fields (which are likely to be irrelevant to your purpose).
We don't currently know when the JSON report will appear within the Web version of EPPI-Reviewer, it's in our plans, but not already explicitly scheduled for a planned release - sorry.
Should the JSON report not suit your needs, an alternative strategy which might even be easier is to export the coding via the new "Quick question report" available from the Web version (available as a sub-menu in the "References" tab, below "coding report"). This provides a very flexible way to export only the coding data that you're interested in. Output is an HTML table, with "decorations" designed to make it easy to import it in Excel (and from there, anywhere).
Hopefully, one of the two options will work for you. I hope this helps,
Sergio