The test is purely a heuristics test. They install the product on a virtual machine and they leave it as it is, without updates or internet connection for a period of time (12 Aug 2011 - 20 Aug 2011). During this period of time they gather new samples. So when they test the detection rate the product doesn't have signatures for the new malware and relies on heuristics to achieve a good score. Another fact worth mentioning is that the on-demand test and the retrospective test should be read together (The document even states that the retrospective test is the second part of the on-demand test). So if a product does well in the on-demand test and not so well in the retrospective test the over-all score could still be favorable (from the pdf: "Users should not be afraid if the products have, in a retrospective test, low percentages. If the antivirus software is always kept up to date, it may be able to detect more samples"). Panda Cloud probably relies on detecting new malware as quickly as possible and since it is a cloud av there shouldn't be a delay until the users benefit from the new signatures. However the retrospective test clearly shows that there is room for improvement when it comes to Panda's heuristics and hopefully they'll improve future versions of the product. Don't forget that the Pro version also has Behavioral analysis of running processes... maybe that helps with the detection of new malware... I don't know, the behavioral analysis didn't help avast! too much in this test.
