Methology: I did a search of http://arxiv.org/find/hep-ex/1/abs:+codes/0/1/0/all/0/1?quer... for "codes" in hep-ex entries at arXiv.org and reported the first few papers which actually contained "codes". (It appears that stemming is in place, as some papers only had 'code'.)

http://arxiv.org/pdf/1510.08141.pdf - "The processes are simulated with the pythia and MadGraph 5 Monte Carlo codes" (authors are from Brazil and Switzerland)

http://arxiv.org/pdf/1510.04063.pdf - "They present new challenges to the simulation codes" (author is from Germany)

http://arxiv.org/pdf/1509.00209.pdf - "The codes and setups used for the formulation of these plots are listed below:" (from Italy) (Note the 'and setups'. This is a clear mistake made by a non-native speaker.)

http://arxiv.org/pdf/1508.00589.pdf - "The Cambridge variable MT2 can already be reliably computed with one of several publicly available codes" (3 of the 8 authors have a US address, though only one of those three is a common name for a native English speaker)

http://arxiv.org/pdf/1507.08764.pdf - "a benchmark for the hadronic interaction Monte Carlo simulations codes" (1 of the 10 authors has a US address. That author has a typically English name)

http://arxiv.org/pdf/1507.06706.pdf - "Both codes are available at" (1 US-based author of 5).

I then repeated the same search with 'programs':

http://arxiv.org/pdf/1510.00391.pdf - "Using a fully automated framework based on the FeynRules and MadGraph5.aMC@NLO programs" (1 UK-based author of 5)

http://arxiv.org/pdf/1509.01918.pdf - "The results are compared with the values used in the simulation programs GEANT4 and UNIMOD." (All authors are based in Russia.)

http://arxiv.org/pdf/1508.05895.pdf - "various high-energy physics programs such as Monte Carlo event generators." (1 UK based author of 6)

http://arxiv.org/pdf/1507.00556.pdf - "This requires an interface between the higher-order theory programs and the fast interpolation frameworks" (23 authors, 18 institutions, and 10 institutions in English speaking countries)

http://arxiv.org/pdf/1506.08759.pdf - "The new estimates presented here are based both on simulation programs (GEANT4 libraries) and theoretical calculations" (Authors are from Italy.)

http://arxiv.org/pdf/1504.06469.pdf - "While many fixed-order and parton shower programs allow" (4 authors, 2 based in English speaking countries, one with a common name for a British person)

This is surely not proof of anything, but it does demonstrate that 1) "codes" is use in scientific computing (though it doesn't show the ratio between "codes" vs. "programs" nor its use in fields other than experimental HEP), 2) there's a suggestion that non-native speakers use "codes" more often, but it's not clear at this level, and 3) there are a lot of non-native speakers pushing papers, while I think the 'software-industry people' who publish in English has a much stronger bias towards native English speaking people.

A more thorough analysis would also need to see if the modern scientific sense is historically consistent; perhaps the industry use is newer though wider coinage.

I think it's both.

But of course, I'm not a native English speaker.