Erll how can a jpeg contain zdepth which is a negative floating point value? Ok so at minimum you loose almost all relevant depth info so you'd need to re range the z channel. So even when you get the channels set your second problem is thet the depth jpegs values are incompatible with real z depth which is why even if you set them correctly it wouldnt work.
However its a bit pointless the have theese values stored in 2 separate jpeg files horrors, the jepg is compressed and theres no guarantee that the compression artifacts will match across the files so what you get will be horribly mangled when you ghet the things out.
Anyway you dont really need to use the z data form the z channel (since your faking the zchannel to begin with, horror estimated estimate) you can instead just connect the second image to a defocus or blur node trough a color curves node thet you shape into a V shape