One of the weird critisisms I've heard (and reflected in your first link) is "but the reproduced study had some slight difference". If a slight difference can make an effect disappear then it's not such an interesting or general effect as the original study claimed to show! Why didn't they already test for the effect of slight differences themselves?

You can't evaluate every single slight variation on your testing environment. That's just the nature of the field.

Your last URL is the same as the first.