Could you clarify what problem you're seeing? The 2s saneMinimum only affects how time is distributed across multiple addresses; it should not change the overall deadline. The following example shows a short timeout working as expected:

$ cat ~/test/dial_fail.go
package main

import (
        "net"
        "fmt"
        "time"
)

func main() {
        startTime := time.Now()
        conn, err := net.DialTimeout("tcp", "192.0.2.1:80", 123*time.Millisecond)
        fmt.Printf("%v %v\n", conn, err)
        fmt.Printf("Finished in %v\n", time.Now().Sub(startTime))
}

$ ../bin/go run ~/test/dial_fail.go
<nil> dial tcp 192.0.2.1:80: i/o timeout
Finished in 123.238884ms

Ah, I am able to reproduce a bug in the partialDeadline calculation. When dialing 2 addresses with a 5 second timeout, the target deadline is recomputed for each address, instead of once at the beginning of Dial:

$ ../bin/go run ~/test/dial_fail.go 
partialDeadline=2.499999662s
partialDeadline=4.999999028s
<nil> dial tcp [2001:db8::1]:80: i/o timeout
Finished in 7.500665372s

I introduced the bug between patch sets 3-4 of the code review:
https://go-review.googlesource.com/#/c/8768/3..4/src/net/dial.go

There is actually a second, pre-existing bug: DialTimeout(..., 5s) will allocate 5 seconds to DNS resolution, and another 5 seconds to TCP.

The fix for both is to call d.deadline(...) exactly once, at the beginning of Dial.

CL https://golang.org/cl/12442 mentions this issue.