For something as specific as this, I'd rather add something to net/http/fcgi than add something to the net/http.Request struct.

Any news about this? We use Apache2/FCGI with Kerberos authentication to our Active Directory, and this missing REMOTE_USER is preventing to use Golang in this case.

No news. Are you interested in adding this? I'm not sure whether you're already familiar enough with Go to add this support. If so, it's all yours. Otherwise, I'm happy to review a change from anybody.

(As long as the change doesn't expand the http.Request struct.)

The mechanics here don't seem super complicated. Rather the most contentious aspect is probably the API. Here's a first crack at something that would be added to net/http/fcgi:

// Serve accepts incoming FastCGI connections on the listener l, creating a new goroutine for each. 
// The goroutine reads requests and then calls handler to reply to them.
// If l is nil, Serve accepts connections from os.Stdin. If handler is nil, http.DefaultServeMux is used.
func ServeWithCgiVariables(l net.Listener, handler Handler) error // 1.

// Handler responds to incoming requests which contain http data and cgi environment variables.
type Handler interface { // 2
    ServeHttpCgiParams(http.ResponseWriter, *Request) // 3.
}

// Request includes an http request and cgi environment variables that were 
// set at the time of the specific request.
type Request struct {
    HTTPRequest http.Request
    CGIEnvVars map[CGIEnvVariable]string // 4.
}

// CGIEnvVariable is a variable set 
type CGIEnvVariable string
const (
    REMOTE_USER CGI EnvVariable = "REMOTE_USER"
    //... 5.
)

Some of the documentation is redundant, but I include it here to illustrate the intention of code that I've left out.
Some of the documentation is left out, but is discussed below:

1. This is not a great function name but we can't overload in go.
2. I'm keeping the same name as the Handler in net package. I thought this was okay to do in go because names are package-prefixed anyway.
3. I package the http and the params in one Request struct instead of having two separate params because I see them as part of the "input" to the handler so we can keep the spirit of the http.Handler with Serve*(output, input).
4. Although this issue's focus is just one of the env variables, probably good to be more general.
5. One of the biggest things this code/documentation leaves out is which variables are being included. Should it be (a) all environment variables or (b) just variables that the code doesn't make a best effort of including in the request? I'm not sure which of these it should be. In either case, the documentation should specify that a best effort is made to include the param values in the http.Request.

(I'd like to work on this issue.)

Thank you for your effort. Could you post anywhere a working version of the code, so we can test it?

For additional motivation: an SO post shows that a resolution to this exact problem has been in demand since 2015.

CL https://golang.org/cl/40012 mentions this issue.