Unless, it is a blocker, it does not get fixed. There are other go projects with its own set of failing tests. It would be more beneficial to fix more failing tests at once.

Keeping this bug opened but not guarantee any progress on it.

Meaning, it is better to comment out the test rather than fix it.

I mean the question should be what is more likely wrong - golang-googlecode-text itself, the test case or even the compiler?

The TestLinking compares size of two compiled binaries. First binary comes from this source code (e.g. as main1.go):

package main
import (
        "fmt"
        "golang.org/x/text/currency"
)
func main() {
        fmt.Print(currency.CLDRVersion)
}

The second one from (e.g. as main2.go):

package main
import (
        "fmt"
        "golang.org/x/text/currency"
)
func main() {
        fmt.Print(currency.Symbol(currency.USD))
}

Compiling the code on x86_64 with golang-1.6.2-1.fc25.x86_64 I have:
main1: 2971752
main2: 2992416

Evaluated condition from currency_test.go:154 that must hold:
size(main2) - size(main1) >= 2*1024


For x86_64 case: 2992416 - 2971752 = 20664 > 2048

Compiling the code on s390x with golang-1.7-0.rc1.fc25.s390x I have:
main1: 2391609
main2: 2392511

The difference:
2392511 - 2391609 = 902 < 2048

Answering your question, the test case is not applicable on s390x.

rawhide build: http://koji.fedoraproject.org/koji/buildinfo?buildID=785820
f25 build: http://koji.fedoraproject.org/koji/buildinfo?buildID=785821

s390 f25 build = http://s390.koji.fedoraproject.org/koji/buildinfo?buildID=420823